Skip to main content
D
Dr. Eugine
Participant
November 25, 2020
Question

Subtraction of a track with environmental noise from the sample

  • November 25, 2020
  • 1 reply
  • 241 views

Forgive if the question was posted and answered already; I am new to audition and I am not an audio technician.

I am analysing sounds generated by precision mechanics for medical implantable instruments, that are running into an anechoic box. The sounds are recorded with a mono mic inside the anechoic box. There is slight environmental noise and we can't imporve the noise insulation.

Today I run a background track and I subtract it manually in Excel.

I would like to know if I can use a second track, recording environmnetal noise by an external mic that I'll use to subtract to the engine sound track. This would be ideal as it is "cleaning" the original track. 

This topic has been closed for replies.

1 reply

SteveG_AudioMasters_
Community Expert
SteveG_AudioMasters_Community Expert
Community Expert
November 25, 2020

No, that won't work, because the noise inside and outside the box won't correlate. Unless there is sample-accurate correlation, you can't do a simple subtraction. The solution (and it's the reason that it exists) is to use Audition's process-based noise reduction. This uses a Fourier-based transform to remove (or at least reduce) noise statistically, and if applied with care it works remarkably well. Almost certainly you will have to increase the FFT size from its default value, probably to the maximum, and be aware that the process works better if you take off small amounts of noise and use multiple passes, resampling the remaining noise each time for the next pass.

 

And yes, this is familiar ground! What you may be able to do is to take a sample of the noise from outside the box and use that as a noise-print for what's inside the box - although generally it works best with the actual noise you are trying to remove - ie, the noise inside the box. It may well be worth experimenting, though.

    D
    Dr. EugineAuthor
    Participant
    November 26, 2020

    SteveG

    this is very helpful, I will surely give a try. Just one clarification: what do you mean by "Unless there is sample-accurate correlation, you can't do a simple subtraction" can you elaborate or make an example of what the correlation consists?

     

    Also let me give fwe more detials about what I am trying to achieve. I am comparing several options of the same device to identify which device gives the lowest perceived noise

    When I run the test I have ancillary tools that generate noises only when the test is running- And those are adding to the background noise. Because it's a comparative among options I don't really care about the ancillary noises (always the same and repeatable)  but external noises are random and I can't compare. I can halt a trial if I hear something but that won't work for ultra/infrasonics. So here the idea to have an external mic, (L channel inside and R-Channel outside) recording simultanously during the trial.

     

    Anyway with the proposal you mentioned (to take a sample of the noise from outside the box and use that as a noise-print) do you mean to take one sample for all trials or a sample for each trial, to be more relevant?

    I will soon try and post my experience

    thank you

    SteveG_AudioMasters_
    Community Expert
    SteveG_AudioMasters_Community Expert
    Community Expert
    November 26, 2020

    Let's do these in order: Sample-accurate means absolutely identical - that's the only way you could do a subtraction. The moment it isn't sample-accurate the inversion will be inaccurate, and therefore the technique won't work. In the real world, it is very, very rare that you can do this, although there are a few instances where as a technique it is useful, although it is very rarely used in its 'pure' form. The most obvious one is with sum and difference recordings with noise in the side channel. Here you can convert from X-Y stereo to sum and difference, treat the noise on just the side (difference) channel and then restore it to X-Y. The conversions are subtractive, but in this instance you are establishing a difference signal - by cancelling just the parts of the signal that are identical in each mic. And this only works if the mics are set up together in the same acoustic space in the first place. 

     

     If your ancilliary noises are always the same and repeatable, you should be able to use a recording of just those as a noise print, and because this is a statistical FFT-based reduction you can store and use this print for any trial using the same equipment and the same location. If this really hasn't altered, then you don't need to take another print  -with the exception of needing to if you do multiple passes, of course. This is quite a normal process; for instance I use it with pipe organ recordings where the blower noise is a constant - just record the blower noise on its own and use that to remove just that noise from the recording. And you can do this with any fixed and constant noise, as long as you do it the way I suggested in my first reply.

     

    But with two different mics in separate locations, you wouldn't even get close to meeting the conditions necessary for a straight subtraction, I'm afraid. This is the reason that audio noise reduction has evolved the way it has!

      Terms & ConditionsAccessibility statement
      Using the community Terms of use Privacy Cookie preferences Do not sell or share my personal information ad choicesAdChoices